Cornell Movie Dialog Chatbot

Introduction

The Cornell Movie Dialog Chatbot is an implementation of a chatbot using NMT - Neural Machine Translation (seq2seq).

The code for the creation of the chatbot has been forked form https://github.com/daniel-kukiela/nmt-chatbot. Once this was done I had to simply follow the setup instructions to get my chatbot up and running. The chatbot here has been trained on the Cornell Movie Dialog corpus using a laptop with rather modest specs (16 GB ram, 2GB V graphics and 1TB storage).

Parameters

Following were the values assigned to the parameters/hyperparameters:

• Vocabulary size : 10,000
• Attention mechanism : Bahdanau
• Number of steps : 50,000
• Number of layers : 2
• Number of units : 512
• Optimizer : Adam
• Encoder type : Bi-directional
• Beam width : 10
• Batch size : 128
• Dropout : 0.2

Getting started

Steps to setup the chatbot are as follows:

1. Download all files in the link https://drive.google.com/open?id=1kZQVW3ZWug-GdnN19QXbQazntAkgwQVg and store it in a folder named 'model'.
2. $ git clone https://github.com/ajoseph12/nmt-chatbot.git
3. Move earlier created folder 'model' into the nmt-chatbot repo.
4. $ pip install -r requirements.txt
5. $ cd model
6. Open the checkpoint file and modify the path to the one where the checkpoint files currently are.
7. Open the hparams file and modify the key values of "src_vocab_file" and "tgt_vocab_file" to the path to which each respective file belongs. ('filepath'\data for scr and tgt vocab )
8. $ cd ..
9. $ python modded-inference.py

Have fun :)